Search

Tree species diversity is vital for maintaining ecosystem functions, yet our ability to map the distribution of tree diversity is limited due to difficulties in traditional field-based approaches. Recent developments in spaceborne remote sensing provide unprecedented opportunities to map and monitor tree diversity more efficiently. Here we built partial least squares regression models using the multispectral surface reflectance acquired by Sentinel-2 satellites and the inventory data from 74 subtropical forest plots to predict canopy tree diversity in a national natural reserve in eastern China. In particular, we evaluated the underappreciated roles of the practical definition of forest canopy and phenological variation in predicting tree diversity by testing three different definitions of canopy trees and comparing models built using satellite imagery of different seasons. Our best models explained 42%-63% variations in observed diversities in cross-validation tests, with higher explanation power for diversity indices that are more sensitive to abundant species. The models built using imageries from early spring and late autumn showed consistently better fits than those built using data from other seasons, highlighting the significant role of transitional phenology in remotely sensing plant diversity. Our results suggested that the cumulative diameter (60%-80%) of the biggest trees is a better way to define the canopy layer than using the subjective fixeddiameter-threshold (5-12 cm) or the cumulative basal area (90%-95%) of the biggest trees. Remarkably, these approaches resulted in contrasting diversity maps that call attention to canopy structure in remote sensing of tree diversity. This study demonstrates the potential of mapping and monitoring tree diversity using the Sentinal-2 data in species-rich forests. [ABSTRACT FROM AUTHOR]